Previous studies concerning the association between physical activity (PA) and mortality in breast cancer yielded mixed results. We investigated the association by performing a meta-analysis of all available studies. Relevant studies were identified by searching PubMed and EMBASE to January 2014. We calculated the summary relative risk (RR) and 95 % confidence intervals (CIs) using random-effects models. The dose-response relationship was assessed by restricted cubic spline model and multivariate random-effect meta-regression. Sixteen cohort studies involving 42,602 patients of breast cancer were selected for meta-analysis. The analyses showed that patients who participated in any amount of PA before diagnosis had a RR of 0.82 (95 % CI 0.74-0.91) for breast cancer-specific mortality (vs. low PA). Those who participated in high PA and moderate PA before diagnosis had a RR of breast cancer-specific mortality of 0.81 (95 % CI 0.72-0.90) and 0.83 (95 % CI 0.73-0.94), respectively. Similar inverse associations of prediagnosis PA were found for all-cause mortality. Postdiagnosis PA on breast cancer-specific and all-cause mortality also showed the same results. Stratifying by body mass index (<25 vs. ≥25) or menopausal status, all the subgroups experienced benefits with PA, with a stronger mortality reduction among overweight women than normal weight women and among postmenopausal women than premenopausal women. A linear and significant dose-response association was only found for breast cancer-specific or all-cause mortality and prediagnosis PA (P for nonlinearity = 0.07 and 0.10, respectively). In conclusion, both prediagnosis and postdiagnosis PA were associated with reduced breast cancer-specific mortality and all-cause mortality.